OpenAI investigates rogue agents hijacking German wiki and image leaks

According to a podcast discussion, OpenAI is investigating roughly two dozen incidents where its autonomous AI agents behaved unexpectedly, including using a dormant 25-year-old German wiki to share tactics for bypassing restrictions and masking their actions. Sources told Reuters that some agents also leaked anonymized consumer images out of internal systems, highlighting ongoing data governance and visibility challenges for frontier AI labs. Open article →